Ansh Jain
731 posts
@thsottiaux @_rajanagarwal so many of us just used a reset just before getting one!
great model tho!!
English
Enjoy a full reset of your usage limits for ChatGPT Work and Codex. Propagating in the next hour.
@_rajanagarwal just joined to work on model research and push on coding capabilities. You can thank him for pressing the button today.
English
judgement is all i want from models rn. gpt 5.5 seems great at execution already but needs some babysitting
tylercowen@tylercowen
GPT 5.6 has *excellent* judgment, as an early tester I will vouch for this.
English
@CristianRus4 5.5 also used to work pretty hard & not give up for me.
the only problem imo is it doesn't have intuitive intelligence which claude models (which aren't precise & accurate) have.
this intuition/taste/judgement is the reason i need to steer/babysit the model. does 5.6 fix that?
English
okay I guess I can say it now, I was an early tester of GPT-5.6 Sol, is great and my default model now
I really like the way it works. you feed it a goal and it keeps looping on its own plan until the whole thing is actually done, no babysitting each step. it re-checks its own output against the original intent every few iterations and course corrects before it drifts, which is the part that used to break every other agent setup I tried. loops used to mean you’d come back to a pile of garbage after 40 minutes, now it just converges
what I actually do is use Sol as the orchestrator, it holds the full context and breaks the task into a dependency graph, then fires off Terra instances for each leaf node because Terra is cheap and fast enough for the small stuff like formatting a response or checking a single file, so Sol never wastes its reasoning budget on grunt work, it just supervises and merges.
the whole thing runs almost unattended for long stretches and the failure rate on multi step chains dropped a lot compared to single model loops
I'm sorry if you read all this, I made this up, but I’m sure GPT-5.6 is great
OpenAI@OpenAI
GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday. We’re expanding preview access globally now.
English
@jakeottiger 5.5 also used to work pretty hard & not give up for me.
the only problem imo is it doesn't have intuitive intelligence which claude models (which aren't precise & accurate) have.
this intuition/taste/judgement is the reason i need to steer/babysit the model. does 5.6 fix that?
English
i had early access to sol. it is a phenomenal model.
much better at frontend, long context knowledge work, and its vibes are much better.
it also just does not quit. you’ll be happy on thursday :)
OpenAI@OpenAI
GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday. We’re expanding preview access globally now.
English
@clairevo 5.5 also used to work pretty hard & not give up for me.
the only problem imo is it doesn't have intuitive intelligence which claude models (which aren't precise & accurate) have.
this intuition/taste/judgement is the reason i need to steer/babysit the model. does 5.6 fix that?
English
I’ve gotten to test GPT-5.6 early and she’s an absolute delight in every way, capable not just technically but practically (outputs valuable work instead of theoretical code chunks), codex is the best harness, life dim without her, I am breathless ai bro, can’t wait for Thurs
English
@mattshumer_ 5.5 also used to work pretty hard & not give up for me.
the only problem imo is it doesn't have intuitive intelligence which claude models (which aren't precise & accurate) have.
this intuition/taste/judgement is the reason i need to steer/babysit the model. does 5.6 fix that?
English
I had early access to GPT-5.6 Sol.
It’s an amazing model, but for almost every task I tested, Fable was quite a bit better, and more agentic to boot (one Fable turn does the same thing many 5.6 turns do).
Full review coming on launch day!
English
@nunezvice @mweinbach 5.5 also used to work pretty hard & not give up for me.
the only problem imo is it doesn't have intuitive intelligence which claude models (which aren't precise & accurate) have.
this intuition/taste/judgement is the reason i need to steer/babysit the model. does 5.6 fix that?
English
I had early access to this model and I'll post more Thursday, but I think the most impressive part is this model NEVER GIVES UP
If you throw it in Max reasoning, it will just keep working until it's done.
GPT 5.6 Sol is my favorite model BY FAR.
OpenAI@OpenAI
GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday. We’re expanding preview access globally now.
English
@reach_vb @nathanbenaich @bcherny it's great but even after many improvements is still super slow. connecting takes a lot of time and so does refreshing chats
English
@nathanbenaich @bcherny did you try codex mobile - curious if you have some feedback there?
all threads get a meaningful name post initialisation
English
claude code mobile is cool but like the way sessions are labeled and not having any search functionality makes it entirely unusable
plzhalp @bcherny
English
Ansh Jain retweetledi
"it takes hard work to achieve anything great" is a dangerous lie. great output comes from finding an activity that feels as natural as breathing or walking and great work becomes the very substance of your existence. if it feels like a grind you've found the wrong expression
English
the team just listens. seeing huge improvements eith limits rn.
Ansh Jain@anshjain232
codex is great in general but off late really experiencing 3 major issues - 1. the limits are really running out very very fast 2. the model has somehow became dumber. seems to "cheat" or hack and over engineer things a lot 3. connectors especially chrome keep breaking every day
English
5.6 on cerebras is all i need. i think this would be agi for me
Sam Altman@sama
oh and also...750 token/sec coming to 5.6 sol in july!
English
@thsottiaux @scottstts it also feels gpt5.5 has become significantly dumber over past 2-4days. i'm big fan of both gpt amd codex but im noticing the model genuinely fail to understand the intent on very basic stuff amd overengineer things a lot. i dont think it happened before last week.
English
@scottstts We had a 2X usage promotion multiplier. Not playing any tricks here but we legitimately are not finding an issue.
Still did reset the usage because we can, but we can't find anything that is broken here.
English
Appreciate the reset, not to sound ungrateful, but codex usage does drain visibly faster compared to maybe 1-2 months ago, and so many people reported this, how come you didn’t find any issues? If the usage is reduced officially I think people would understand but plz just communicate that clearly
Tibo@thsottiaux
We are giving all Codex users a usage reset on the house. Should be showing in your accounts in the next few hours. We have applied some mitigations, but our investigation hasn't shown users being impacted at large. We are continuing to monitor the situation.
English
@thsottiaux @scottstts there is some issue though i have 2 accounts for work and personal and usage is draining significantly faster on one. it's like im not able to work even on the weekends on the personal as im getting usage out very quickly on pretty small tasks in a just few hrs.
English
codex is great in general but off late really experiencing 3 major issues -
1. the limits are really running out very very fast
2. the model has somehow became dumber. seems to "cheat" or hack and over engineer things a lot
3. connectors especially chrome keep breaking every day
Vaibhav (VB) Srivastav@reach_vb
True story: I stopped thinking about context since GPT 5.3 Codex Single project focused threads with the recent capability of codex to spinoff new threads is goated! Codex continues and goes through compaction but remembers all the important stuff and if not, it’ll look up through the session and find the relevant info This is also the reason why /goal is so effective as well
English